dect
/
asterisk
Archived
13
0
Fork 0
This repository has been archived on 2022-02-17. You can view files and clone it, but cannot push or open issues or pull requests.
asterisk/utils/hashtest.c

411 lines
9.6 KiB
C
Raw Normal View History

This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
/*
* Asterisk -- An open source telephony toolkit.
*
* Copyright (C) 2007, Steve Murphy
*
* Steve Murphy <murf@digium.com>
*
* See http://www.asterisk.org for more information about
* the Asterisk project. Please do not directly contact
* any of the maintainers of this project for assistance;
* the project provides a web site, mailing lists and IRC
* channels for your use.
*
* This program is free software, distributed under the terms of
* the GNU General Public License Version 2. See the LICENSE file
* at the top of the source tree.
*/
/*! \file
*
* \brief A program to thoroughly thrash a hash table, testing
* out locking safety, and making sure all functionality
* is functioning. Run with 5 or more threads to get that
* fully intense firestorm of activity. If your
* hash tables don't crash, lock up, or go weird, it must
* be good code! Even features some global counters
* that will get slightly behind because they aren't lock-protected.
*
* \author Steve Murphy <murf@digium.com>
*/
/*** MODULEINFO
<support_level>extended</support_level>
***/
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
#include "asterisk.h"
ASTERISK_FILE_VERSION(__FILE__, "$Revision$")
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
#include <pthread.h>
#include <sys/stat.h>
#include <signal.h>
#include <errno.h>
#include "asterisk/lock.h"
#include "asterisk/hashtab.h"
#include "asterisk/channel.h"
#include "asterisk/utils.h"
#include "asterisk/module.h"
int testno = 1;
/* stuff we need to make this work with the hashtab stuff */
#if !defined(LOW_MEMORY)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
int64_t ast_mark(int prof_id, int x)
{
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
}
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
void pbx_substitute_variables_helper_full(struct ast_channel *c, struct varshead *headp, const char *cp1, char *cp2, int cp2_size, size_t *used);
void pbx_substitute_variables_helper_full(struct ast_channel *c, struct varshead *headp, const char *cp1, char *cp2, int cp2_size, size_t *used)
{
}
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
struct ht_element
{
char *key;
char *val;
};
static int hashtab_compare_strings_nocase(const void *a, const void *b)
{
const struct ht_element *ae = a, *be = b;
return ast_hashtab_compare_strings_nocase(ae->key, be->key);
}
#if 0
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
static int hashtab_compare_strings(const void *a, const void *b)
{
const struct ht_element *ae = a, *be = b;
return ast_hashtab_compare_strings(ae->key, be->key);
}
static unsigned int hashtab_hash_string(const void *obj)
{
const struct ht_element *o = obj;
return ast_hashtab_hash_string((const void *)o->key);
}
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
static unsigned int hashtab_hash_string_nocase(const void *obj)
{
const struct ht_element *o = obj;
return ast_hashtab_hash_string_nocase((const void*)o->key);
}
/* random numbers */
static int my_rand(int incl_low, int incl_high, unsigned int *seedp)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
{
if (incl_high == 0)
return 0;
return incl_low + (rand_r(seedp) % incl_high);
}
/* the testing routines */
static int glob_highwater = 0;
struct ast_hashtab *glob_hashtab = 0;
unsigned int glob_seed = 0;
int els_removed = 0;
int els_added = 0;
int els_lookedup = 0;
int els_found = 0;
int els_traversals = 0;
/* all the operations to perform on the hashtab */
static void add_element(void)
{
char keybuf[100];
struct ht_element *x = malloc(sizeof(struct ht_element));
sprintf(keybuf,"key%08d", glob_highwater++);
x->key = strdup(keybuf);
x->val = strdup("interesting data");
ast_hashtab_insert_immediate(glob_hashtab, x);
els_added++;
}
static void traverse_elements(void)
{
struct ht_element *el;
int c=0;
struct ast_hashtab_iter *it = ast_hashtab_start_write_traversal(glob_hashtab);
#ifdef DEBUG
printf("Traverse hashtab\n");
#endif
while ((el = ast_hashtab_next(it))) {
c++;
}
ast_hashtab_end_traversal(it);
els_traversals++; /* unprotected, sometimes off, but, not really important, either */
}
static void * del_element(unsigned int *seedp)
{
char keybuf[100];
struct ht_element *el, lookup;
int x;
/* pick a random element from 0 to highwater-1 */
x = my_rand(0,glob_highwater-1,seedp);
sprintf(keybuf, "key%08d", x);
#ifdef DEBUG
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
printf("Removing %s", keybuf);
#endif
lookup.key = keybuf;
el = ast_hashtab_remove_object_via_lookup(glob_hashtab, &lookup);
if (el) {
#ifdef DEBUG
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
printf("...YES (el=%x)\n", (unsigned long)el);
#endif
free(el->key);
free(el->val);
free(el);
els_removed++;
} else {
#ifdef DEBUG
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
printf("...NO.\n");
#endif
return 0;
}
return el;
}
static int lookup_element(unsigned int *seedp)
{
char keybuf[100];
struct ht_element *el, lookup;
int x;
/* pick a random element from 0 to highwater-1 */
x = my_rand(0,glob_highwater-1,seedp);
sprintf(keybuf, "key%08d", x);
lookup.key = keybuf;
el = ast_hashtab_lookup(glob_hashtab, &lookup);
els_lookedup++;
if (el)
els_found++;
if (el)
return 1;
return 0;
}
static void *hashtest(void *data)
{
int my_els_removed = 0;
int my_els_added = 0;
int my_els_lookedup = 0;
int my_els_found = 0;
int my_els_traversals = 0;
int my_testno = testno++;
int its;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
/* data will be a random number == use as a seed for random numbers */
unsigned long seed = (unsigned long)data;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
printf("hashtest thread created... test beginning\n");
/* main test routine-- a global hashtab exists, pound it like crazy */
for(its=0;its<100000;its++)
{
void *seed2 = &seed;
int op = my_rand(0,100, seed2);
if (op<60) {
my_els_lookedup++;
#ifdef DEBUG
printf("%d[%d]: LOOKUP\n", my_testno, its);
#endif
if ((my_els_lookedup%1000)==0) {
printf(".");
fflush(stdout);
}
if (lookup_element(seed2))
my_els_found++;
} else if (op < 61) { /* make this 61 and it'll take 15 minutes to run */
#ifdef DEBUG
printf("%d[%d]: TRAVERSE\n", my_testno, its);
#endif
traverse_elements();
my_els_traversals++;
} else if (op < 80) {
#ifdef DEBUG
printf("%d[%d]: REMOVE\n", my_testno, its);
#endif
if (del_element(seed2))
my_els_removed++;
} else {
my_els_added++;
#ifdef DEBUG
printf("%d[%d]: ADD\n", my_testno, its);
#endif
add_element();
}
}
printf("\nhashtest thread %d exiting.... lookups=%d/%d, added=%d, removed=%d, traversals=%d;\n",
my_testno, my_els_found, my_els_lookedup, my_els_added, my_els_removed, my_els_traversals);
printf("\ntotals..................... lookups=%d/%d, added=%d, removed=%d, traversals=%d;\n",
els_found, els_lookedup, els_added, els_removed,els_traversals);
pthread_exit(0);
}
static void run_hashtest(int numthr)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
{
pthread_t thr[numthr];
void *thrres[numthr];
int i, biggest, resize_cnt, numobjs, numbuckets;
/* init a single global hashtab, then... */
glob_hashtab = ast_hashtab_create(180000, hashtab_compare_strings_nocase, ast_hashtab_resize_java, ast_hashtab_newsize_java, hashtab_hash_string_nocase, 1);
printf("starting with %d elements in the hashtable...\n", ast_hashtab_capacity(glob_hashtab));
/* set a random seed */
glob_seed = (unsigned int)time(0);
srand(glob_seed);
/* create threads, each running hashtest */
for(i=0;i<numthr;i++)
{
unsigned long z = rand();
printf("starting hashtest thread %d....\n",i+1);
if (ast_pthread_create(&thr[i], NULL, hashtest, (void*)z)) {
printf("Sorry, couldn't create thread #%d\n", i+1);
}
printf("hashtest thread spawned.... \n");
}
/* collect threads, each running hashtest */
for(i=0;i<numthr;i++)
{
printf("waiting for thread %d....\n", i+1);
if (pthread_join(thr[i], &thrres[i])) {
printf("Sorry, couldn't join thread #%d\n", i+1);
}
printf("hashtest thread %d done.... \n",i+1);
}
/* user has to kill/intr the process to stop the test? */
ast_hashtab_get_stats(glob_hashtab, &biggest, &resize_cnt, &numobjs, &numbuckets);
printf("Some stats: longest bucket chain: %d; number of resizes: %d; number of objects: %d; capacity: %d\n",
biggest, resize_cnt, numobjs, numbuckets);
}
int main(int argc,char **argv)
{
if (argc < 2 || argc > 2 || atoi(argv[1]) < 1)
{
printf("Usage: hashtest <number of threads>\n");
exit(1);
}
/* one arg == number of threads to create */
run_hashtest(atoi(argv[1]));
return 0;
}
#if !defined(LOW_MEMORY)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
int ast_add_profile(const char *x, uint64_t scale)
{
return 0;
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
}
#endif
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
int ast_loader_register(int (*updater)(void))
{
return 1;
}
int ast_loader_unregister(int (*updater)(void))
{
return 1;
}
void ast_module_register(const struct ast_module_info *x)
{
}
void ast_module_unregister(const struct ast_module_info *x)
{
}
void ast_register_file_version(const char *file, const char *version);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
void ast_register_file_version(const char *file, const char *version)
{
}
void ast_unregister_file_version(const char *file);
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
void ast_unregister_file_version(const char *file)
{
}
void ast_log(int level, const char *file, int line, const char *function, const char *fmt, ...)
{
va_list vars;
va_start(vars,fmt);
printf("LOG: lev:%d file:%s line:%d func: %s ",
level, file, line, function);
vprintf(fmt, vars);
fflush(stdout);
va_end(vars);
}
void __ast_verbose(const char *file, int line, const char *func, const char *fmt, ...)
This is the perhaps the biggest, boldest, most daring change I've ever committed to trunk. Forgive me in advance any disruption this may cause, and please, report any problems via the bugtracker. The upside is that this can speed up large dialplans by 20 times (or more). Context, extension, and priority matching are all fairly constant-time searches. I introduce here my hashtables (hashtabs), and a regression for them. I would have used the ast_obj2 tables, but mine are resizeable, and don't need the object destruction capability. The hashtab stuff is well tested and stable. I introduce a data structure, a trie, for extension pattern matching, in which knowledge of all patterns is accumulated, and all matches can be found via a single traversal of the tree. This is per-context. The trie is formed on the first lookup attempt, and stored in the context for future lookups. Destruction routines are in place for hashtabs and the pattern match trie. You can see the contents of the pattern match trie by using the 'dialplan show' cli command when 'core set debug' has been done to put it in debug mode. The pattern tree traversal only traverses those parts of the tree that are interesting. It uses a scoreboard sort of approach to find the best match. The speed of the traversal is more a function of the length of the pattern than the number of patterns in the tree. The tree also contains the CID matching patterns. See the source code comments for details on how everything works. I believe the approach general enough that any issues that might come up involving fine points in the pattern matching algorithm, can be solved by just tweaking things. We shall see. The current pattern matcher is fairly involved, and replicating every nuance of it is difficult. If you find and report problems, I will try to resolve than as quickly as I can. The trie and hashtabs are added to the existing context and exten structs, and none of the old machinery has been removed for the sake of the multitude of functions that use them. In the future, we can (maybe) weed out the linked lists and save some space. git-svn-id: http://svn.digium.com/svn/asterisk/trunk@89129 f38db490-d61c-443f-a65b-d21fe96a405b
2007-11-09 16:00:22 +00:00
{
va_list vars;
va_start(vars,fmt);
printf("VERBOSE: ");
vprintf(fmt, vars);
fflush(stdout);
va_end(vars);
}
void ast_register_thread(char *name)
{
}
void ast_unregister_thread(void *id)
{
}
#ifdef HAVE_BKTR
struct ast_bt *ast_bt_create(void);
struct ast_bt *ast_bt_create(void)
{
return NULL;
}
int ast_bt_get_addresses(struct ast_bt *bt);
int ast_bt_get_addresses(struct ast_bt *bt)
{
return 0;
}
char **ast_bt_get_symbols(void **addresses, size_t num_frames);
char **ast_bt_get_symbols(void **addresses, size_t num_frames)
{
char **foo = calloc(num_frames, sizeof(char *) + 1);
if (foo) {
int i;
for (i = 0; i < num_frames; i++) {
foo[i] = (char *) foo + sizeof(char *) * num_frames;
}
}
return foo;
}
void *ast_bt_destroy(struct ast_bt *bt);
void *ast_bt_destroy(struct ast_bt *bt)
{
return NULL;
}
#endif